Ayedris/amharic-ecommerce-ner

10Academy Week-4 Challenge # Amharic E-commerce Data Extractor for FinTech Insights This project is part of the 10Academy Week 4 AI Mastery Challenge (June 18–24, 2025). The goal is to develop a system for extracting structured insights from Telegram-based e-commerce channels in Amharic, enabling EthioMart to centralize listings and evaluate vendors for micro-lending opportunities. ## Project Objectives - Scrape and preprocess Amharic messages from Telegram e-commerce channels. - Label data using CoNLL format for NER. - Fine-tune transformer-based NER models (XLM-R, Amharic-BERT, etc.). - Compare models using metrics like F1-score and interpretability tools (SHAP, LIME). - Score vendors using extracted data and Telegram metadata for loan eligibility. ### Project Structure ``` amharic-ecommerce-ner/ │ ├── data/ │ ├── raw/ │ └── labeled/ │ ├── notebooks/ │ ├── data_ingestion.ipynb │ ├── labeling_sample.ipynb │ ├── ner_training.ipynb │ └── model_comparison.ipynb │ ├── scripts/ │ ├── telegram_scraper.py │ ├── preprocess.py │ ├── label_converter.py │ ├── train_ner.py │ ├── interpretability.py │ └── vendor_scorecard.py │ ├── models/ │ ├── outputs/ │ ├── interim_summary.pdf │ └── final_report.pdf │ ├── README.md ├── .gitignore ├── requirements.txt ├── dvc.yaml └── LICENSE ```